AI accelerates validation, but it does not create market evidence
An assistant can generate ideas, segments, objections and proposals in minutes. This broadens exploration, but it also makes it easy to build a convincing story without speaking to a single customer: validation needs real behaviour. Validating means reducing uncertainty through real behaviour, not accumulating plausible text.
Use AI as an analyst and sparring partner: to formulate better questions, look for contradictions and synthesise signals. Reserve decisions for verifiable evidence and contact with the market.
If the only person who has confirmed the problem is an AI, you have not validated anything yet.
Formulate a testable hypothesis
Write: “We believe that [segment] has [specific problem] in [situation], which causes [consequence]. If we offer [proposal], we expect [measurable behaviour].” Separate problem, solution, channel, price and viability: each element contains a different uncertainty.
Ask AI to act as a sceptical customer, competitor and finance lead. Request hidden assumptions and reasons why the hypothesis could be false. Then turn the relevant criticisms into research questions.
Research the problem before presenting the solution
- How the person currently solves the situation.
- When it last happened.
- The time, money or risk it costs them.
- Which alternatives they have tried and why they failed.
- Who decides, who uses and who pays.
- What would have to change for them to adopt another solution.
Interview people in the segment and avoid hypothetical questions such as “would you use it?” Ask about recent facts. AI can transcribe and group patterns, but review quotations and context before drawing conclusions.
Distinguish signal from courtesy
Verbal interest is a weak signal. Actions such as providing data, introducing a decision-maker, booking a trial, signing a letter of intent or paying are stronger. Define in advance what evidence you need to move forward.
Look for disconfirmation. If ten interviews repeat the problem but nobody changes their behaviour, urgency may be low or the current alternative may be sufficient.
Design the cheapest experiment
- Landing page with a proposal and a real call to action.
- Manually delivered concierge service.
- Clickable prototype that validates the flow.
- Wizard of Oz: an automated experience with human operation behind it and appropriate transparency.
- Pre-sale, paid trial or limited commitment.
The goal is to learn before building. Measure conversion by segment, acquisition cost, activation, repeat use and willingness to pay. Do not confuse traffic with interest or free sign-ups with sustainable demand.
Check whether AI is the core or an accessory
Ask whether the proposal would lose its advantage without AI, whether it improves with usage data and whether it solves a limitation of scale, personalisation or knowledge. If a conventional rule or automation delivers the same result, it may be the more profitable option.
When AI is necessary, also validate minimum quality, cost per interaction, latency, privacy and oversight capacity. A desirable proposal may not be viable with the available margin.
Decide with a learning log
- Hypothesis and initial confidence level.
- Experiment and success criterion.
- Observed result, including contradictory data.
- What you have learned and what changes.
- Decision: persevere, modify, pivot or stop.
Keep versions so that you do not reinterpret results later. This discipline prevents AI’s speed from becoming a succession of ideas without accumulated learning.
From validated problem to MVP
When there is evidence of a problem and behaviour, build the minimum experience that tests the proposal. Link validation to a use-case prioritisation matrix and define from the start how you will measure value, quality and risk.
Discovery interview template
- Tell me about the last time the problem happened.
- What were you trying to achieve and what did you do step by step?
- Where did the greatest difficulty occur?
- What impact did it have in terms of time, money or risk?
- Which alternative did you use and what did it cost you?
- Who else was involved in the decision?
- What would have to happen for you to look for another solution?
Do not present your idea until you understand the current behaviour. At the end, summarise what you understood and ask for corrections. Then label patterns, but preserve fragments and cases that contradict the hypothesis.
Frequently asked questions about validation
How many interviews are needed?
Enough to observe patterns within a segment, not to achieve statistical significance. Start with five, adjust the questions and continue until new information decreases. Do not mix segments to reach a number.
Can AI simulate customers?
It can help you rehearse questions and explore objections, but it does not replace real people. A model reproduces training information and your instructions; it does not demonstrate urgency or willingness to pay.
When should you pivot?
When the evidence invalidates a core hypothesis but reveals a better problem, segment or mechanism. Changing only the message without recording learning is not a pivot.
What we learned while validating the I3OS embryo
The origin of I3OS was an SEO pilot applied to real work, not an idea validated only with a demo. Iterating with clients allowed us to check which parts of the method were reusable and discover opportunities in other areas of Impulsa3. That experience confirms this article’s rule: AI can accelerate exploration, but the decision to move forward needs behaviour, results and recorded learning.
If you need to validate a business idea or new service through rapid experiments and real evidence, Impulsa3 can help you design the process and turn learning into a viable solution.
